pandas iat[] — read or write one cell by position

What it does

df.iat[r, c] reads or writes exactly one cell by integer position — labels, slices, and masks are all off the table. df.iat[0, 1] hands back a single scalar (an np.float64 under pandas 3), and df.iat[0, 1] = 25.0 writes straight into the parent frame. Negative positions work: df.iat[-1, 0] is last row, first column. Series join the party one axis smaller — s.iat[1] reads, s.iat[1] = 99 writes. Anything beyond a scalar pair raises 'iAt based indexing can only have integer indexers'; a position past the end raises IndexError — unlike .at, iat never enlarges the frame.

Why it matters

The workhorse scenario: you computed WHERE to fix (row position from argmax or a loop counter), and you want to fix THAT spot without dragging labels around. iat is the shortest honest write in modern pandas — no chained-assignment gray zones under Copy-on-Write, no dtype guessing. It is also the fastest single-cell read in the indexing family: ~9.5 µs per lookup on a 100k-row frame versus ~13.7 µs for .iloc and ~19 µs for .loc (measured on pandas 3.0.5 this morning). The discipline is the hidden feature: if your 'iat' call contains a list or a variable that might be a label, you are in the wrong method — and pandas tells you immediately instead of returning something subtly wrong.

Example

$ df = pd.DataFrame({'sensor': ['RX-1', 'RX-2', 'TX-4', 'TX-9'],
...     'temp_c': [21.4, 19.8, 23.1, 18.6],
...     'batt':   [92, 87, 55, 99]},
...     index=['a1', 'a2', 'a3', 'a4'])
>>> df.iat[0, 1]

np.float64(21.4)

$ hot = df.temp_c.to_numpy().argmax()   # -> 2: the hottest sensor is row 2
>>> df.iat[hot, 1] = 25.0              # re-check confirmed: overwrite it
>>> df.temp_c.to_list()

[21.4, 19.8, 25.0, 18.6]

$ s = pd.Series([10, 20, 30], index=['x', 'y', 'z'])
>>> s.iat[1] = 99
>>> s.to_list()

[10, 99, 30]

$ df.iat[[0, 1], 0]

ValueError: iAt based indexing can only have integer indexers

$ df.iat['a3', 1]

ValueError: iAt based indexing can only have integer indexers

$ df.iat[4, 1]

IndexError: index 4 is out of bounds for axis 0 with size 4

$ df.iat[-1, 0]

'TX-9'

Common flags

df.iat[r, c]

read one cell by row/column POSITION — pure ints, scalar out

df.iat[r, c] = x

write one cell in the parent frame — the CoW-proof single-cell write by position

df.iat[-1, 0]

negative positions work, numpy-style: -1 is the last row

s.iat[i]

Series form — one integer, one element (same for reads and writes)

df.iat[[0, 1], 0]

ValueError 'can only have integer indexers' — lists and masks belong to .iloc

df.iat['a3', 1]

ValueError — labels belong to .at, never iat (the most common iat mistake)

df.at['a3', 'temp_c']

the label twin — identical scalar contract, first hash-resolves the label before the same final read